October 2026 | Strategy

Beyond the AI Pilot: How Executives Lead AI Transformation

Companies can produce impressive AI pilots faster than they can turn them into durable business results. The software may work as intended while the initiative stalls because workflows, incentives, decision rights, and accountability have not changed around it. That gap raises the questions executives ultimately have to answer: Is AI solving the right business problem? What must change before it can scale? And which decisions should leaders refuse to delegate?

Kartik Hosanagar, Wharton professor and faculty co-director of Wharton Human-AI Research, studies how AI reshapes firms, markets, and managerial decision making. In this Q&A, he discusses how executives can move beyond isolated use cases toward business-model change, stronger judgment, and accountable AI leadership. Wharton Executive Education adds program context for leaders who want to explore these questions in greater depth.

Wharton@Work: How should executives approach change management for AI transformation?

Kartik Hosanagar: A mistake many companies often make is to treat the model or the tool as the change. A team gets access to a powerful system, runs a promising pilot, and assumes the organization is on its way. But a capable model is only one component of a working business system.

The harder issues are around redesigning workflows and reskilling the team. Consider workflows. Where does the tool enter the workflow? Which handoffs should disappear, and which new ones should appear? Who can act on its output? Who monitors and reviews exceptions? What happens when the system is confidently wrong? If every decision still has to pass through the old approval chain, technical capability will not translate into much organizational value. I often say that “You can make every employee 30 percent more productive without making the company 30 percent more productive.”

I find it useful to think of AI transformation in four stages. The first stage is AI as a tool for individual productivity: an employee uses AI to draft, summarize, or analyze. The second is workflow redesign, where the organization changes how a piece of work moves from start to finish. The third is AI agents with decision rights that can carry out a sequence of tasks within defined limits. The deepest change occurs when AI becomes part of the operating model: how the firm allocates work, information, authority, and accountability.

Many companies are still measuring success at the first stage. An employee completes a task faster, so the pilot looks productive. Yet the surrounding process may not move any faster at all. The work may simply wait longer for review because AI creates more output than any team can absorb or shifts human effort from production to verification. As I said, local productivity and system productivity are different things.

That is why the executive role is less about selecting a model or tool and more about redesigning the conditions under which people and models work together. Leaders responsible for cross-functional adoption need to understand the technology, but they also need to think through incentives, workflows, governance, and change.

I have discussed the urgency of moving from experimentation to enterprise change in an earlier Wharton@Work article and in recent thought pieces.

Wharton@Work: How is generative AI changing business models, and where can executives learn with peers?

KH: In terms of AI’s impact on business models, I would start with the constraints that shape the current business model. What does AI make abundant? What remains scarce? Where might value move as a result? Most early business cases start with labor savings because those are easier to estimate. That is reasonable, but it can be strategically narrow.

When the cost of producing analysis, software, images, or customer communications falls, execution may become less scarce. The scarce elements may then be judgment, proprietary context, customer trust, coordination across functions, or the willingness to own a decision. A company that only automates the old process may miss that shift. It is using a new capability to preserve an old business model.

Take management consulting for example. The primary role of consultants used to be to help clients tackle difficult business questions by researching industries, analyzing problems, and developing recommendations. Today, AI can perform much of that research and generate plausible answers to well-defined questions. As answers become abundant, more value may shift to asking the right questions and taking greater ownership of the answers. That means helping clients implement recommendations, not simply delivering them, and potentially tying fees more closely to outcomes. I believe consulting will need to move in this direction because a business model built primarily around producing answers will become harder to sustain when plausible answers are abundant.

This does not mean every industry will be overturned or that every company needs an entirely new model. AI can make one constraint less binding while exposing another. A firm may be able to create many more product concepts, for example, only to discover that testing, distribution, brand credibility, or organizational buy-in now limits growth. The strategic question is not simply, “What can we generate?” It is, “What bottleneck appears next?”

Leaders should treat new offerings and operating models as hypotheses. Identify where value is created today, ask which assumption AI changes, and test the alternative. Then watch for the operational cost that the spreadsheet left out.

Wharton@Work: Do CMOs need a marketing-specific AI program or a general AI leadership program?

KH: A marketing-specific program makes sense when the CMO’s decisions concern the customer journey, brand, search, pricing, creative work, or marketing measurement. If the mandate is to redesign the enterprise operating model or set governance across business units, a general AI leadership program is the better fit.

Marketing also has a distinctive issue: AI agents may become a new audience or customer type, rather than merely another channel for reaching one. Agents may research products, filter alternatives, recommend a choice, and eventually transact on a person’s behalf. Adoption and behavior will vary, so marketers should not assume that every purchase will soon be delegated. But the direction raises a different set of questions.

How does a brand become legible and credible to an agent? Which product facts must be structured so a machine can compare them? What happens to persuasion when an intermediary screens out much of the message before a person sees it? A marketer may need to communicate with both a human buyer and a system acting under that buyer’s preferences.
That changes more than media planning. It touches product information, distribution, customer data, reputation, and the design of the customer experience. A CMO leading company-wide adoption would need to widen the lens, because the hardest constraints may sit outside marketing.

W@W: Generative AI has made AI feel accessible. Do executives still need data and analytics fundamentals?

KH: More than before, I would argue. Generative AI arrived through a conversational interface, so it can feel as though the analytical layer underneath has been abstracted away. But many consequential business decisions are still prediction and measurement problems. Which customers are likely to leave? What will demand look like next quarter? Did the change we made cause the improvement, or did something else happen at the same time? A fluent answer to a poorly framed measurement question is still the wrong answer. Leaders need to interpret data, distinguish correlation from causation, design a useful experiment, and judge whether an apparent improvement is real. They do not need to become data scientists, but they need enough understanding to question an analysis, especially when that analysis can appear fluent without being valid.

This shows up most clearly when an executive has to evaluate a claim. Let’s say that someone reports that engagement rose after a new AI-driven recommendation feature launched. A leader who understands experimentation asks what the comparison group was, whether the customer mix changed, and how long the effect persisted. A leader without that grounding can only accept the number or reject it on instinct. Neither is a sound decision process.

So, I would not think of analytics as the previous generation of this work and AI as the current one. The analytical foundation allows an executive to question the analysis and distinguish a real gain from an artifact of measurement. This skillset is difficult to build in a short burst. It develops when a leader learns a concept, applies it to something consequential in the organization, and returns with better questions. A program that supports this learning process and requires participants to work with their own business context can provide that kind of sustained application.

W@W: Do AI certificate programs build practical executive capability?

KH: I would separate the credential from the capability. A certificate can document that someone completed a defined course of study. The more important question is whether the learning changed that person's ability to leverage AI in their work.

Evidence of capability appears in what changed in the learner’s ability. Can the executive frame a business problem before reaching for a tool, question an output, spot a missing assumption, and decide when expert review is necessary? Can the learning transfer to a consequential decision at work?

Foundational knowledge still matters. There is a temptation to say that executives no longer need to understand a domain because AI can explain it or perform parts of the work. But without a mental model of the subject, people have a weak basis for verifying the answer. They may accept a polished response that omits a key stakeholder, mistakes correlation for causation, or recommends something that cannot be implemented.

Practical capability therefore requires concepts to be used and tested. Completing a course documents a course of study; it is not proof that someone can make every consequential AI decision well. What matters afterward is whether the learner can use the foundation, test it against real work, and recognize when an answer exceeds their ability to verify it. In short, I would look for programs that teach the fundamentals well and then help executives apply them to real business problems. That combination of foundational understanding and practical application is what my Wharton faculty colleagues and I aim to build into our AI programs.

W@W: What are the trade-offs between online and in-person AI programs, and how should executives choose?

KH: Online programs are often a good choice for building foundational knowledge, especially for executives who want flexibility. Self-paced formats make it easier to learn around a full-time work schedule, and they are typically less expensive than residential programs. They can also work well for targeted skill development in areas such as AI strategy, machine learning, or digital transformation.

The challenge is that AI is moving unusually fast. In a self-paced program, recorded material can become dated even when the underlying concepts remain useful. For that reason, I would complement an online program with more current sources, such as podcasts, newsletters, or Substacks that track recent developments in AI.

In-person programs offer a different set of advantages. The curriculum can be updated much closer to the date of delivery. In the programs I teach, for example, we revisit and redesign parts of the curriculum before each session because the technology and business landscape are changing so quickly. In-person programs also create more room for discussion, peer learning, and relationship-building. Executives meet others who are wrestling with similar questions about generative AI, and those conversations can be as valuable as the formal content. Many participants in our programs stay in touch afterward.

The trade-offs are time, cost, and flexibility. In-person programs require executives to step away from work, travel, and commit to a fixed schedule, and they are generally more expensive. Online programs offer much more scheduling freedom and can be a better fit for someone primarily seeking a strong foundation or a specific set of skills.

So, I would choose based on the objective. If the goal is to build AI literacy efficiently and on your own schedule, a strong online program may be enough, especially if you supplement it with current material. If the goal is to work through strategic questions, learn from other senior leaders, and develop a network of peers facing similar challenges, the added investment in an in-person program can be worthwhile.

W@W: Where can executives build judgment about responsible AI and ethical decision making online?

KH: AI presents a massive opportunity to organizations. But poorly governed AI also represents a significant reputational, litigation, and regulatory risk. Organizations should design guardrails, test AI systems more systematically, route exceptions for review, and preserve the option for a human override.

Human involvement and exceptional handling offer one solution but they have to be operationally meaningful. Reviewing every output may help in some settings, but it does not scale to every application. In addition, the form of oversight should match the risk. A low-stakes drafting tool and a system influencing consequential decisions do not call for the same level of scrutiny.

Overall, boards and senior teams will increasingly need to anticipate legal, ethical, and business risks posed by AI. They can explore these ideas in "AI on the Agenda: Three Tactical Strategies for Boards."

In terms of how Wharton can help leaders prepare for ethical governance of AI, we were early in assessing AI’s opportunities and risks. For example, I proposed an AI governance framework called “AI Bill of Rights” back in 2019. My colleague, Professor Kevin Werbach, founded and runs the Accountable AI Lab at Wharton. There are many opportunities to partner with and learn from Wharton faculty.

W@W: How can executives evaluate whether an AI model or tool solves the right business problem?

KH: Start with the decision, not the model. What decision or process are you trying to improve? How is that decision made today? What does the current process cost, how well does it perform, and what would happen if you did nothing? Without that baseline and counterfactual, a model can look impressive while contributing very little.

Model accuracy is relevant, but it is rarely the whole evaluation. Suppose a system produces a good recommendation most of the time. Is the recommendation complete enough to act on? Are the failures visible, or do they look deceptively plausible? Does the workflow route high-risk cases to someone with the expertise and authority to intervene? Does the system improve the outcome that matters, or merely make one intermediate task faster?

Executives should also account for integration and failure costs. A model may save time on routine cases but create expensive errors in the tail. It may require so much review that the apparent gain disappears. Or it may generate a strong first draft that still leaves a manager responsible for omitted stakeholders, questionable assumptions, and the consequences of implementation. Those are product questions as much as technical ones.

The practical test is whether the full system makes a consequential decision better than the present alternative, with failures the organization can detect and contain. Executives evaluating a course or program should look for the same discipline in its cases and exercises: Does it require them to identify the decision, establish a baseline, examine failure modes, and assign ownership of the outcome? If the team cannot do those things, it is too early to declare that the model solves the problem.

Links:

• Generative AI and Business Transformation
• Leading AI and Digital Transformation
• Business Model Innovation in the Age of AI
• AI in Marketing: Creating Customer Value in an AI-Driven Enterprise
• Leadership Program in AI
• Analytics for Strategic Growth: AI, Smart Data, and Customer Insights
• Digital Strategy in the Era of AI Certificate
• AI for Business
• Strategies for Accountable AI

---

Share This

Subscribe to the Wharton@Work RSS Feed